Constructing a 22-year internal wave dataset for the northern South China Sea: spatiotemporal analysis using MODIS imagery and deep learning DOI Creative Commons
Xudong Zhang, Xiaofeng Li

Earth system science data, Journal Year: 2024, Volume and Issue: 16(11), P. 5131 - 5144

Published: Nov. 6, 2024

Abstract. Internal waves (IWs) are an important ocean phenomenon facilitating energy transfer between multiscale processes. Understanding such processes necessitates the collection and analysis of extensive observational data. IWs predominantly occur in marginal seas, with South China Sea (SCS) being one most active regions, characterized by frequent large-amplitude IW activities. In this study, we present a comprehensive dataset for northern SCS (https://doi.org/10.12157/IOCAS.20240409.001, Zhang Li, 2024), covering area from 112.40 to 121.32° E 18.32 23.19° N, spanning period 2000 2022 250 m spatial resolution. During 22 years, total 15 830 MODIS images were downloaded further processing. Out these, 3085 high-resolution true-color identified contain information included precise positions extracted using advanced deep learning techniques. categorized into four regions based on distributions. This classification enables detailed analyses characteristics, including their temporal distributions across entire its specific sub-regions. Interestingly, our reveals characteristic “double-peak” patterns aligned lunar day, highlighting strong connection tidal cycles. Furthermore, identifies two quiescent zones within clusters influenced underwater topography, regional variations characteristics suggesting underlying mechanisms which merit investigation. There also three gap distinct clusters, may indicate different sources. The constructed holds significant potential studying IW–environment interactions, developing monitoring prediction models, validating numerical simulations, serving as educational resource promote awareness interest research.

Language: Английский

Reconstructing 3-D Thermohaline Structures for Mesoscale Eddies Using Satellite Observations and Deep Learning DOI
Yingjie Liu, Haoyu Wang, Fei Jiang

et al.

IEEE Transactions on Geoscience and Remote Sensing, Journal Year: 2024, Volume and Issue: 62, P. 1 - 16

Published: Jan. 1, 2024

Mesoscale eddies are circular water currents found widely in the ocean and significantly impact ocean's circulation, distribution, biology. However, our comprehension of eddies' three-dimensional (3D) structures remains constrained due to scarcity in-situ data. Therefore, we introduce a novel deep learning model, 3D-EddyNet, designed for reconstructing 3D thermohaline structure mesoscale eddies. Utilizing multi-source satellite data Argo profiles collected from North Pacific Ocean between 2000 2015, optimized 3D-EddyNet model by adjusting image sizes, introducing Convolutional Block Attention Module, incorporating eddy physical parameters. Results demonstrate remarkable accuracy, with an average root mean square error (RMSE) 0.32 °C (0.03 psu) temperature (salinity) within anticyclonic 0.41 (0.04 cyclonic upper 1000 m. We applied reconstruct Kuroshio Extension (KE) Oyashio Current (OC) regions, demonstrating its capability accurately represent both vertically horizontally. The consistency averaged ARMOR3D dataset KE OC regions underscores robust generalizability indicating model's ability infer when unavailable. distinctive advantage offered enhances understand dynamics, overcoming challenges posed limited availability

Language: Английский

Citations

11
Deep learning techniques for enhanced sea-ice types classification in the Beaufort Sea via SAR imagery DOI Creative Commons
Yan Huang, Yibin Ren, Xiaofeng Li

et al.

Remote Sensing of Environment, Journal Year: 2024, Volume and Issue: 308, P. 114204 - 114204

Published: May 13, 2024

This study proposes a dual-branch encoder U-Net (DBU-Net) deep learning model to classify sea ice types based on synthetic aperture radar (SAR) images in the Beaufort Sea. The DBU-Net can segment multi-year (MYI), first-year (FYI), open water (OW), and leads SAR images. We design fuse polarization grey-level co-occurrence matrix (GLCM) information of improve model's classification capability. is subsequently fine-tuned using lead samples identify leads. 24 Sentinel-1 acquired Sea are utilized for training testing. accuracy (Acc), mean intersection over union (mIoU), kappa coefficient (Kappa) employed as evaluation metrics. Experiments show that achieves 91.83%/0.841/0.849 Acc/mIoU/Kappa classifying MYI, FYI, OW, significantly outperforming three traditional models support vector machine, random forest, or convolutional neural network. Compared with original U-Net, GLCMs 1.45%/4.4%/2.8% OW. metrics detection 99.49%/0.801/0.754. Besides, 454 fed into optimal generate 80 m products winters 2018–2022. As MYI draws wide attention FYI complementary area during Winter, we discuss variation generated explore relationship between MYI's High. found export 2018/19 winter was due large summer remains abnormal motion caused by southeast shifting Atmospheric Pressure High (Beaufort High). import 2020/21 strong northward powerful

Language: Английский

Citations

9
Tropical cyclone intensity forecasting using model knowledge guided deep learning model DOI Creative Commons
Chong Wang, Xiaofeng Li, Gang Zheng

et al.

Environmental Research Letters, Journal Year: 2024, Volume and Issue: 19(2), P. 024006 - 024006

Published: Jan. 8, 2024

Abstract This paper developed a deep learning (DL) model for forecasting tropical cyclone (TC) intensity in the Northwest Pacific. A dataset containing 20 533 synchronized and collocated samples was assembled, which included ERA5 reanalysis data as well satellite infrared (IR) imagery, covering period from 1979 to 2021. The u -, v - w -components of wind, sea surface temperature, IR historical TC information were selected inputs. Then, TC-intensity-forecast-fusion (TCIF-fusion) developed, two special branches designed learn multi-factor forecast 24 h intensity. Finally, heatmaps capturing model’s insights are generated applied original input data, creating an enhanced set that results more accurate forecasting. Employing this refined input, (model knowledge) used guide TCIF-fusion modeling, model-knowledge-guided achieved error 3.56 m s −1 Pacific TCs spanning 2020–2021. show performance our method is significantly better than official subjective prediction advanced DL methods by 4% 22%. Additionally, compared operational approaches, model-guided knowledge can landfalling TCs.

Language: Английский

Citations

6
Deep learning solver unites SDGSAT-1 observations and Navier–Stokes theory for oceanic vortex streets DOI
He Gao, Baoxiang Huang, Ge Chen

et al.

Remote Sensing of Environment, Journal Year: 2024, Volume and Issue: 315, P. 114425 - 114425

Published: Sept. 24, 2024

Language: Английский

Citations

6
Applications of knowledge distillation in remote sensing: A survey DOI
Yassine Himeur, Nour Aburaed, Omar Elharrouss

et al.

Information Fusion, Journal Year: 2024, Volume and Issue: unknown, P. 102742 - 102742

Published: Oct. 1, 2024

Language: Английский

Citations

4
Integrating GIS-Remote Sensing: A Comprehensive Approach to Predict Oceanographic Health and Coastal Dynamics DOI
Ramesh Krishnamoorthy, Kazuaki Tanaka, M. Amina Begum

et al.

Remote Sensing in Earth Systems Sciences, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 6, 2025

Language: Английский

Citations

0
Advancing forecasting capabilities: A contrastive learning model for forecasting tropical cyclone rapid intensification DOI Creative Commons
Chong Wang, Nan Yang, Xiaofeng Li

et al.

Proceedings of the National Academy of Sciences, Journal Year: 2025, Volume and Issue: 122(4)

Published: Jan. 21, 2025

Tropical cyclones (TCs), particularly those that rapidly intensify (RI), pose a significant threat due to the uncertainty in forecasting them. RI TC periods, which by at least 13 m/s within 24 h, remain challenging forecast accurately. Existing models achieve probability of detection (POD) 82.6% and false alarm rate (FARate) 27.2%. To address this, we developed contrastive-based (RITCF-contrastive) model, utilizing satellite infrared imagery alongside atmospheric oceanic data. The RITCF-contrastive model was tested on 1,149 periods Northwest Pacific from 2020 2021, achieving POD 92.3% FARate 8.9%. improves previous addressing sample imbalance incorporating structural features, leading 11.7% improvement 3 times reduction compared existing deep learning methods. not only enhances but also offers unique approach these dangerous weather events.

Language: Английский

Citations

0
Estimating ocean currents from the joint reconstruction of absolute dynamic topography and sea surface temperature through deep learning algorithms DOI Creative Commons
Daniele Ciani, Claudia Fanelli, Bruno Buongiorno Nardelli

et al.

Ocean science, Journal Year: 2025, Volume and Issue: 21(1), P. 199 - 216

Published: Jan. 27, 2025

Abstract. Our study focuses on absolute dynamic topography (ADT) and sea surface temperature (SST) mapping from satellite observations, with the primary objective of improving satellite-derived ADT (and derived geostrophic currents) spatial resolution. Retrieving consistent high-resolution SST information space is challenging, due to instrument limitations, sampling constraints, degradations introduced by interpolation algorithms used obtain gap-free (L4) analyses. To address these issues, we developed tested different deep learning methodologies, specifically convolutional neural network (CNN) models that were originally proposed for single-image super Building upon recent findings, conduct an Observing System Simulation Experiment (OSSE) relying Copernicus numerical model outputs (with respective temporal resolutions 1 d 1/24°), present a strategy further refinements. Previous OSSEs combined low-resolution L4 equivalent ADTs “perfectly known” SSTs derive dynamical features. Here, introduce realistic processing errors modify concurrently predict synthetic, products. This modification allows us evaluate potential enhancement in while integrating constraints through tailored, physics-informed loss functions. The networks are thus trained using OSSE data subsequently applied Marine Service SSTs, allowing reconstruct super-resolved currents at same spatiotemporal resolution employed OSSE. A 12-year-long time series (2008–2019) presented validated against situ-measured drogued drifting buoys via spectral suggests CNNs beneficial standard altimetry mapping: they generally sharpen gradients, consequent correction direction intensities respect altimeter-derived investigation focused Mediterranean Sea, quite challenging region its small Rossby deformation radius (around 10 km).

Language: Английский

Citations

0
Internal solitary waves in the Banda Sea, a pathway between Indian and Pacific oceans: Satellite observations and physics-AI hybrid forecasting DOI
Xudong Zhang, Haoyu Wang, Xiaofeng Li

et al.

Remote Sensing of Environment, Journal Year: 2025, Volume and Issue: 323, P. 114733 - 114733

Published: April 3, 2025

Language: Английский

Citations

0
Attention-enhanced deep learning approach for marine heatwave forecasting DOI
Yiyun Liu, Le Gao, Shuguo Yang

et al.

Acta Oceanologica Sinica, Journal Year: 2025, Volume and Issue: unknown

Published: April 15, 2025

Language: Английский

Citations

0